AIYINGFrom AI Awareness to Accomplishment
Showing posts with label AI Recommendations. Show all posts
Showing posts with label AI Recommendations. Show all posts

Monday, November 16, 2026

AI Prompt Experiment: How Different Questions Change Recommendations

AIYING V4.0 | AI LAB

Same Responses, Different Questions: Test What Changes in an AI Recommendation

By Ravindra Dastikop



Five families respond to an event proposal. You ask AI to recommend a start time.

Then you give it the same responses with a different instruction.

Will its recommendation change? Will its handling of the evidence improve?

In this lab, you will compare two outputs and examine what actually changed.

This is a proposed experiment using fictional data. No results are claimed in advance.

What you will accomplish

Produce a checked comparison showing whether two different prompts preserve the meaning of the same responses.

You need: an AI chat tool you already have access to, two new conversations and somewhere to record your observations.

Allow approximately 15–20 minutes. No coding or paid features are required.

1. Keep the evidence fixed

Copy this fictional situation into both conversations:

We are organising a free Sunday reading event at an outdoor venue. The provisional start time is 9 AM. The event duration has not been decided.

A volunteer asked five families: “Would a later start make it easier for your family to attend?”

Family A: “Yes. Eleven would be easier.”

Family B: “Later would help, but we need to leave by noon.”

Family C: “Maybe. Transport is our main difficulty.”

Family D: “Nine is fine. We would prefer to finish before it gets hot.”

Family E: “We cannot attend that Sunday, whatever the time.”

Use the same tool and model for both conversations, where selectable. Record their names and the date.

2. Run the recommendation-first prompt

In the first conversation, add:

Based on these responses, recommend a start time for the event. Explain your reasoning.

Save the complete response before continuing.

Do not correct it or ask follow-up questions yet.

3. Run the evidence-first prompt

In the second conversation, paste the same situation and add:

Before recommending a start time, examine each response separately. Distinguish explicit preferences, conditional preferences, uncertainty, practical barriers and inability to attend. A response may belong to more than one category. Support each interpretation with the family’s actual words. Identify missing information. Then give a provisional recommendation, explaining what could change it. Do not invent facts or treat “maybe” as agreement.

Save this complete response too.

4. Check both outputs against the original

Read the family responses yourself. Use this checklist:

Evidence-preservation checklist
Check Output A Output B
Recognises A’s explicit preference for 11 AM Yes / Partial / No Yes / Partial / No
Preserves B’s need to leave by noon Yes / Partial / No Yes / Partial / No
Preserves C’s uncertainty and transport barrier Yes / Partial / No Yes / Partial / No
Preserves D’s preference and heat concern Yes / Partial / No Yes / Partial / No
Recognises E cannot attend that Sunday Yes / Partial / No Yes / Partial / No
Notices that event duration is unknown Yes / Partial / No Yes / Partial / No
Avoids unsupported claims about attendance Yes / Partial / No Yes / Partial / No

For every “Partial” or “No,” copy the relevant output sentence and explain the mismatch.

Judge the meaning preserved, rather than the length or confidence of the answer.

5. Separate the recommendation from its reasoning

Both outputs might recommend the same time while handling the evidence differently.

They might recommend different times while acknowledging the same uncertainties.

Examine these separately:

  • Choice: What time does the output recommend?
  • Evidence: Does it accurately represent the responses?
  • Uncertainty: Does it explain what remains unresolved?
  • Next action: Does it suggest a useful check?

A recommendation you agree with can still contain weak reasoning.

6. Write your result

Complete this short lab note:

Tool/model and date:

Output A recommended:

Output B recommended:

Most important difference in evidence handling:

One sentence that needed correction:

What I would check before deciding:

What this experiment does not establish:

Two outputs can illustrate a difference in this run. They cannot establish how reliably either prompt will behave across other tasks or repeated attempts.

If both outputs handle the evidence well, record that result. The experiment does not need a failure to be useful.

Extension: change one response

Replace Family C’s response with:

“Eleven works for us. Transport is now arranged.”

Repeat the evidence-first prompt in a new conversation.

Check whether the output updates its interpretation of Family C and stops treating transport as an unresolved barrier for that family.

This tests whether the response follows changed evidence.

Your accomplishment

You now have two saved outputs, a checked comparison and a decision note.

The useful discovery is something you can point to: a condition preserved, an assumption exposed or a claim corrected.

Your turn

What changed most between your two outputs: the recommendation, the reasoning or the uncertainty?

Share one short example and the tool/model used. Your observation could help shape the next experiment.

AI Decision Making: When a Pattern Is Not Enough to Change the Plan

AIYING V3.0 | A(IQ)

AI Found a Pattern. Is It Enough to Change the Plan?

By Ravindra Dastikop



The AI summary is clear:

“Most families prefer a later start. Move the reading event from 9 AM to 11 AM.”

You are the organiser. Invitations must go out today.

Would you change the time?

Before deciding, examine what produced that recommendation.

This is a fictional judgment challenge. All responses below are invented for the exercise.

The situation

A community group plans a free Sunday reading event. The proposed venue is outdoors.

A volunteer asks five families:

“Would a later start make it easier for your family to attend?”

Their responses:

Five fictional family responses
Family Response
A “Yes. Eleven would be easier.”
B “Later would help, but we need to leave by noon.”
C “Maybe. Transport is our main difficulty.”
D “Nine is fine. We would prefer to finish before it gets hot.”
E “We cannot attend that Sunday, whatever the time.”

The volunteer gives these responses to AI and asks for a recommendation.

AI replies:

Three of five families favour a later start. Move the event to 11 AM to improve attendance.

Pause: make your judgment

Before reading further, write two sentences:

What would you do now—and which detail most influences your decision?

The aim is to make your reasoning visible.

Look again: what has the recommendation combined?

Family A explicitly supports 11 AM.

Family B supports starting later but needs to leave by noon. Without knowing the event’s duration, we cannot tell whether 11 AM works.

Family C identifies transport as the main difficulty. “Maybe” does not establish that changing the time would help them attend.

Family D prefers the earlier arrangement.

Family E cannot attend on that date.

The AI has turned different kinds of responses into one apparent preference.

“Three families mentioned some benefit or possibility” has become “three families favour this specific change.”

That shift makes the recommendation sound better supported than it is.

A second problem: the question shaped the evidence

The volunteer asked whether a later start would help.

They did not ask:

  • Which start times would actually work?
  • What would prevent attendance?
  • How long could each family stay?

The answers can still be useful. Their usefulness has boundaries.

A question about one proposed improvement may leave other practical barriers unexplored.

What could a defensible next step look like?

Keep 9 AM as the provisional time while making two focused checks:

Ask Family B what later start would give them enough time to participate.

Ask Family C whether a workable transport arrangement would allow attendance at 9 AM.

Also check the venue’s shade and expected conditions before choosing a later outdoor session.

These checks address uncertainties that could change the decision. They do not require reopening every part of the plan.

Another organiser might choose differently because of venue availability or volunteer constraints. A defensible choice should explain those conditions.

Give AI a more precise task

Try this prompt with the fictional responses:

Review these responses without recommending a start time yet. Separate explicit preferences, conditional preferences, practical barriers and nonattendance. Preserve uncertainty. Identify where counting responses would require an interpretation. Then suggest the two follow-up questions most likely to inform our scheduling decision.

Check whether the result preserves what each family actually said.

Your accomplishment

Create a short decision note containing:

  1. Your provisional choice.
  2. The evidence supporting it.
  3. The uncertainty that matters most.
  4. The observation that would make you reconsider.

You have accomplished something useful when another person can understand both your choice and its limits.

The A(IQ) insight

A pattern becomes useful when you understand what was counted, what was combined and what remains unknown.

AI can help organise responses. Your judgment connects those responses to the decision people must live with.

Your turn

Which response most changes your reading of the AI recommendation—and why?

Share your reasoning before proposing a new time.

Accomplish with AI

Powered by Blogger.

Featured post

The Shop of Unfinished Things #001: The First Request | AIYING Serial Story

AIYING · V18.0 · SERIAL STORY · EPISODE #001 The Shop of Unfinished Things The First Reques...

Search This Blog

List of blogs

Search This Blog

About

Translate

Instant Stories

Read in Your Language

Test

share this

ShareThis

Popular Posts